OpenAI has announced the launch of its new language model, “GPT-4.5,” as a research model for ChatGPT Pro users, despite warning that it is not an advanced computing model, raising questions about its capabilities and limitations.
Article Contents:
- Introduction
- Improvements and New Features
- Limitations and Warnings
- Training Methods and Techniques Used
- Interaction and Collaboration Enhancements
- Official Statements and Rollout Plans
- Future Prospects and Upcoming Developments
- Conclusion
- Frequently Asked Questions
Introduction
In a new step within its ongoing quest to develop artificial intelligence technologies, OpenAI announced the launch of the “GPT-4.5” model today. This model is considered the largest and newest release in the language model series, and it has been launched as a research model for ChatGPT Pro users. The company described it as its “most knowledgeable model yet,” while noting that despite improvements to its writing and world knowledge capabilities, it is not an advanced computing model as some might imagine.
Improvements and New Features
The “GPT-4.5” model features several technical improvements compared to previous versions. It exhibits advanced writing capability through enhanced expression style and text fluency, in addition to updated world knowledge that makes it more accurate in providing information. The model’s persona has also been developed to become warmer and more natural during interaction, as it is now capable of recognizing patterns and connecting information in a way that contributes to solving practical problems such as writing and programming. OpenAI affirms that interacting with “GPT-4.5” will become more harmonious with users thanks to these enhancements.
Limitations and Warnings
Despite numerous improvements, OpenAI has warned against considering “GPT-4.5” an advanced or pioneering model in the field of computing. According to reports, the model’s performance falls short of models like o1 and o3-mini in certain research readiness benchmarks. Some references were removed from official documentation after publication, reflecting OpenAI’s commitment to providing an accurate assessment of the model’s capabilities.
Training Methods and Techniques Used
OpenAI relied on new supervision techniques in training “GPT-4.5” alongside traditional methods such as supervised fine-tuning and reinforcement learning from human feedback. Synthetic data generated by the model formerly codenamed “Strawberry”—the codename for the o1 model used in the training process—was also utilized. This approach highlights its ability to reduce error and hallucination rates compared to previous models like “GPT-4o.”
Interaction and Collaboration Enhancements
OpenAI researchers emphasized that “GPT-4.5” has become a more collaborative model with users, delivering precise responses while lowering the rate of false hallucinations. One researcher stated that the model has become warmer and more understanding of users, outperforming “GPT-4o” in most human evaluation metrics. Thus, the model strives to deliver a more natural and emotional conversational experience.
Official Statements and Rollout Plans
OpenAI CEO Sam Altman noted that “GPT-4.5” is a massive and costly model, explaining that it “will not shatter benchmark records,” signaling his expectation that the model represents an incremental improvement rather than a revolution in the AI world. The model is scheduled to begin rolling out to ChatGPT Pro users, followed by Plus and team users next week, and subsequently educational institutions and major enterprises. The model is also now available on the Azure AI Foundry platform alongside new models from other companies.
Future Prospects and Upcoming Developments
According to previous reports published in Notepad magazine, OpenAI had planned to launch “GPT-4.5” at the end of February, with the announcement of “GPT-5” slated for later in May. Sam Altman described “GPT-5” as an integrated system combining multiple technologies, including the o3 model used for complex reasoning processes, as part of the company’s efforts to develop a comprehensive model that could be classified in the future as “artificial general intelligence.” These efforts come in the context of OpenAI’s pursuit to aggregate large language models that contribute to achieving higher computing capabilities and smarter interactions.
Conclusion
The launch of the “GPT-4.5” model is considered a new step in OpenAI’s journey of artificial intelligence development, delivering tangible improvements in writing, knowledge, and user interaction. Despite warnings regarding its status as a non-advanced computing model, the model reflects the company’s commitment to refining its technologies and providing more natural and collaborative experiences for users. With upcoming developments continuing, the question remains as to the extent of these enhancements’ impact on the future of language models and efforts toward achieving artificial general intelligence.
Frequently Asked Questions
1. What is the “GPT-4.5” model?
It is the newest language model released by OpenAI, utilized as a research model for ChatGPT Pro users with enhancements in writing, knowledge, and interaction.
2. What are the main improvements the model offers?
It features improved writing style, updated world knowledge, and a developed persona that becomes warmer and more natural when interacting with users.
3. Why did OpenAI warn against considering the model an advanced model?
The company pointed out that “GPT-4.5” does not deliver revolutionary new capabilities compared to previous versions, and its performance is lower than models like o1 and o3-mini in certain tests.
4. What technologies were used to train the model?
The training relied on new supervision techniques alongside traditional supervised fine-tuning and reinforcement learning from human feedback methods, utilizing synthetic data through the “Strawberry” (o1) model.
5. How does the model’s interaction differ from previous versions?
It has become more collaborative with users, providing precise responses while lowering the incorrect hallucination rate, making it superior in certain human evaluation benchmarks.
6. What are the distribution plans for the model?
The model is rolling out initially to ChatGPT Pro users, then moving to Plus and team users, followed by educational institutions and major enterprises, and it is also available on Azure AI Foundry.
7. What are the future prospects after the launch of “GPT-4.5”?
OpenAI is working on developing the “GPT-5” model, which will combine multiple technologies in a step toward achieving higher capabilities and potentially reaching artificial general intelligence.
